teamai-cli – Tencent's Open-Source AI Team Collaboration CLI Tool

1. What is teamai-cli

TeamAI CLI is an open-source MIT-licensed command-line tool developed by Tencent, designed to uniformly manage a team's AI programming configuration system. This tool centralizes AI collaboration resources such as skills, rules, MCP, and hooks into a Git repository, and distributes the configurations to each team member's local AI tools through a standardized workflow of push → MR → merge → pull. TeamAI CLI supports knowledge accumulation driven by friction signals, team knowledge retrieval, and codebase graph spectrum capabilities, enabling standardized and governable AI collaboration at the team level, and addressing issues of configuration fragmentation and knowledge silos in multi-tool environments.

Technical Positioning and Domain: TeamAI CLI belongs to the AI engineering infrastructure domain, specifically focusing on team-level configuration management and knowledge collaboration for AI programming tools. It resides in the configuration orchestration and distribution layer of the AI programming toolchain, situated between team management needs and individual AI coding tools (such as Claude Code, Cursor, Codex, etc.), acting as a central hub and distribution channel for unified configurations. Its unique positioning lies in introducing the mature Git workflow from the software development domain into AI configuration management.

Development Background: This tool was open-sourced by Tencent, based on its extensive internal AI programming practices. As AI coding tools like Claude Code, Cursor, and Codex become widely adopted in enterprises, teams face management challenges such as scattered configurations, inconsistent standards, and the inability to reuse knowledge. Tencent abstracted an internally validated management solution into a general-purpose tool and open-sourced it, using the permissive MIT license to facilitate community adoption and further development. This reflects Tencent's exploration in the direction of AI engineering.

Core Value: TeamAI CLI addresses the core pain point of configuration fragmentation in multi-tool environments for team AI collaboration. It achieves versioned management and automated distribution of configurations through native Git workflows, ensuring that team coding standards, MCP tools, and security hooks are uniformly pushed to each member's local tools. Additionally, it employs a friction signal detection and knowledge retrieval mechanism to transform individual experience into shared team assets, closing the loop from experience generation to reuse.

Technical Features: The tool is built upon native Git workflows, achieving automatic configuration synchronization through the SessionStart hook, without requiring additional daemons or server-side infrastructure. Its multi-Agent adaptation layer can automatically convert and distribute a single YAML declaration to over 9 mainstream AI programming tools, significantly reducing maintenance costs. The knowledge retrieval mechanism integrates BM25 search with code graph-enhanced ranking, enabling automatic judgment of task relevance.

2. Key Features

• Unified Configuration Management: Centralize the team's skills, rules, MCP, hooks, env, and other resources within a single Git repository, enabling version-controlled and review-based management of AI programming configurations. This feature consolidates configuration files that were previously scattered across individual developers' local environments into a centralized repository, ensuring that configuration changes are auditable, rollable back, and traceable.

• Native Git Distribution Mechanism: Utilize a standardized workflow of push → MR → merge → pull to automatically synchronize team configurations to each member's local AI tools. This mechanism leverages developers' familiar Git operations and uses the SessionStart hook to automatically pull the latest configuration at the start of a session, reducing configuration update delays to within the session startup cycle.

• Multi-Agent Unified Compatibility: Supports 9+ mainstream AI programming tools, including Claude Code, Cursor, Codex, CodeBuddy, WorkBuddy, and Qoder. Administrators maintain a single standard configuration in the TeamAI repository, and the system automatically converts it into native formats for each tool and distributes it to the corresponding directories, eliminating the burden of maintaining multiple configurations for the same team across different tools.

• Team-Level Fine-Grained Control: Achieve precise subscription and filtering based on three dimensions: Roles (job roles), Tags (topic tags), and Sources (source repositories). Different roles can subscribe to different skill sets and rule sets, enabling on-demand configuration distribution rather than full pushes, thereby improving the accuracy of configuration delivery.

• Unified Hosting for Team MCP and Hooks: Declare MCP service configurations in YAML once, and the tool automatically generates native MCP configuration files for different AI tools. The Hooks event system supports injecting custom scripts at event points such as SessionStart and PreToolUse, used for automated controls like security scanning and compliance interception.

• Friction Signal-Driven Knowledge Accumulation: Automatically detect friction signals during session interactions, such as interruptions, refusals, and retries, triggering an experience summarization process. After confirmation by developers, the system automatically removes private information and pushes the lessons learned into the team repository, enabling automatic conversion of individual experiences into team knowledge.

• Pre-Task Knowledge Recall: Automatically search the team knowledge base before initiating an AI task, using the BM25 algorithm combined with code graph-enhanced ranking to determine whether to recall based on relevance to the current task. Automatically skip irrelevant tasks to avoid interference from unrelated context, improving the accuracy and efficiency of AI task execution.

• Codebase Knowledge Graph: Parse the codebase AST (Abstract Syntax Tree) and combine it with heuristic dependency analysis to build a structured dependency graph across repositories. This graph provides semantic association support for knowledge recall, enhancing the AI's understanding of the codebase structure and dependency relationships, thus improving the accuracy of recall results.

Image source: official article

3. How to Use

1. Install the CLI tool: Execute the global installation command npm install -g teamai-cli. Before installation, ensure that your local Node.js environment meets the tool's requirements (recommended Node.js version 18 or higher). After installation, you can verify the success of the installation using the command teamai --version.

2. Create a team shared repository: Set up a team-specific repository on GitHub, GitLab, or a self-hosted Git service to store AI configuration resources. Team members must be granted write access to this repository (required for submitting branches and initiating MRs), and it is recommended to configure branch protection rules to allow merging into the main branch only via Merge Request.

3. Initialize team configuration: The team administrator runs teamai init <repository address> and completes the initialization process by following the prompts, including OAuth login, repository association, and member registration. This step establishes trust between the team repository and the TeamAI CLI and generates the initial directory structure.

4. Members join the project: After team members navigate to their respective project working directories, they run teamai init <repository address> and select either project-level or user-level installation mode based on their needs. Project-level installation binds the configuration to the current project directory, while user-level installation applies the configuration to all projects for the current user.

5. Write team resources: Create or modify configuration files such as skills, rules, docs, mcp, and hooks in the team repository. It is recommended to use the TeamAI CLI's local validation command (e.g., teamai validate) before submitting to check YAML formatting and the validity of configuration items, thereby reducing the cost of MR reviews.

6. Push and review changes: Run teamai push to automatically create a feature branch and initiate a Merge Request for team review. Administrators or designated reviewers approve and merge the MR on the code hosting platform. All configuration changes undergo a review process to ensure configuration quality and traceability of changes.

7. Automatic synchronization and manual updates: Every time a team member starts an AI session, the SessionStart hook automatically triggers teamai pull to fetch the latest configuration into the local AI tool directories. Users can also manually run teamai pull at any time to force synchronization. If configuration synchronization issues occur, run teamai status to check the difference status between local configurations and the repository.

8. Enable knowledge retrieval and accumulation: Run teamai recall enable to deploy a sub-agent for automatic knowledge retrieval before task execution. After experiencing high-friction conversations, run /teamai-share-learnings to automatically anonymize and push the lessons learned into the team repository. Run teamai dashboard to view the web dashboard, or use teamai digest to generate a weekly report for the team.

4. Pros and Cons Analysis

5. Comparative Analysis with Similar Tools

Selection Recommendations: For medium to large R&D teams that have deeply adopted Git collaboration standards, use multiple AI programming tools, and emphasize knowledge accumulation, TeamAI CLI is the more complete infrastructure choice. Its coverage extends from configuration management to knowledge accumulation and data analysis, and the MIT license facilitates internal customization, effectively integrating AI usage into the team's engineering management system. Ruler, with its support for 10+ tools and comprehensive rule coverage for multi-agent systems, is suitable as a lightweight alternative for teams that primarily seek rule unification and have already accumulated a large number of rule files across various tools.

For teams that have already decided to adopt Gemini CLI as their unified tool, or for enterprises with high requirements for data security and policy control, the built-in policy configuration capabilities of Gemini CLI can meet the needs for in-depth control ranging from system prompts to tool permissions. Crush, on the other hand, is more suitable for individual developers or small teams who need to manage prompts and session history but have not yet established a team-level collaboration system. Overall, TeamAI CLI demonstrates a more systematic comprehensive coverage across four dimensions—configuration, knowledge, control, and analysis—making it the ideal choice for long-term team AI collaboration infrastructure construction among similar open source tools.

6. Editor's Summary

From a technological innovation perspective, TeamAI CLI introduces Git's version control and collaboration mechanisms into the field of AI configuration management, offering a management paradigm that aligns with developers' existing workflows. Its friction signal-driven knowledge accumulation mechanism, combined with the BM25+ knowledge retrieval method enhanced by a graph, provides a design approach worth emulating for knowledge management in AI collaboration tools—automating the decoupling of the "experience generation" and "experience organization" processes. The tool's ability to unify MCP and Hooks management also reflects its deep understanding of the AI programming toolchain ecosystem.

From a practical value standpoint, TeamAI CLI directly addresses core pain points in team AI collaboration, such as fragmented configurations, difficulties in knowledge accumulation, and weak control mechanisms. By incorporating configuration changes into MR reviews, the tool brings AI integration behaviors under the team's governance framework. Its knowledge accumulation and retrieval mechanisms transform individual experience into team assets, continuously reducing the cost of repeated trial and error for team members. In terms of target users, this tool is most suitable for development teams with clear Git collaboration standards, who use multiple AI programming tools, and who prioritize knowledge accumulation and security management. It holds significant value for first-line internet companies and mid-to-large technical teams that have a high adoption rate of AI tools.

It should be noted that the effectiveness of TeamAI CLI is closely tied to the maturity of the team's collaboration standards: factors such as Git process standards, MR review systems, and the extent of code repository integration directly impact the tool's performance. As an open-source tool currently in the community growth phase, its ecosystem completeness and industry practice validation cases still require more time to mature. Additionally, the varying levels of compatibility among AI tools with MCP and Hooks will affect the full realization of unified configurations. In the long term, as the adoption of AI programming tools continues to rise in enterprises, team-level AI configuration management will become a necessary component of the R&D infrastructure. TeamAI CLI's exploration in this direction demonstrates a certain level of foresight and reference value.

7. Application Scenarios

• Unified Configuration Management for Multi-Tool Teams: Some team members use Claude Code, others use Cursor, and still others use Codex. The administrator maintains a unified set of skills, rules, and MCP configurations in the TeamAI repository, which all members automatically synchronize with. This scenario ensures that "the same coding standards, the same MCP tools, and the same set of security hooks" are uniformly applied across the team, eliminating collaboration deviations caused by tool differences.

• Rapid AI Onboarding for New Members: After joining the team, new members only need to run teamai init once to automatically receive the team's accumulated skills, rules, and best practices, without needing to manually configure each AI tool individually. This scenario reduces the new member's "AI tuning period" from days to minutes, enabling them to have the same level of AI-assisted capabilities as existing team members before even starting actual coding work.

• Unified Security and Compliance Control: The security team configures PreToolUse hooks in the team repository's hooks/hooks.yaml, injecting security policies such as sensitive information detection and outgoing content scanning. After the configuration passes the MR review, it is automatically distributed to all team members, ensuring that all AI session behaviors are subject to the same unified security policies, preventing security incidents or compliance risks caused by missing personal configurations.

• Automatic Accumulation and Cross-Team Reuse of Troubleshooting Experience: When a team member resolves a complex port conflict issue, the interruptions and retries in the session are captured by friction signals and trigger a knowledge accumulation process. After de-identifying the data, this experience is pushed into the team repository. Thereafter, when other members encounter similar issues, the AI automatically recalls this experience and provides targeted recommendations, forming a closed-loop team knowledge cycle of "troubleshooting → knowledge accumulation → reuse."

• Cross-Team Skill Sharing and Subscription: Team A publishes its tested and validated public skills to GitHub, and Team B subscribes to Team A's skill source using teamai source add, enabling cross-team reuse of best practices within the organization. This scenario is suitable for skill sharing among multiple teams, such as frontend, backend, and operations, within large organizations, avoiding resource waste caused by redundant development of the same skills across teams.

8. FAQ

Q: Which AI coding tools does TeamAI CLI support? Is the level of compatibility consistent across all tools?
A: TeamAI CLI currently supports over 9 mainstream AI coding tools, including Claude Code, Cursor, Codex, CodeBuddy, WorkBuddy, and Qoder. The compatibility varies by tool: for tools that fully support the MCP protocol and the Hooks event mechanism (e.g., Claude Code), TeamAI CLI can activate all its features; for tools with limited native support, there may be a degradation in MCP or Hooks functionality. For specific compatibility status, please refer to the support matrix documentation in the official repository.

Q: What are the core advantages of TeamAI CLI compared to manually writing configuration files locally?
A: The core advantages lie in three areas: version management, collaborative distribution, and knowledge accumulation. Manually writing configuration files locally cannot achieve historical traceability of versions, change reviews, or collaborative editing by multiple people. TeamAI CLI integrates configuration into the Git workflow, automatically distributing all changes to the entire team after they are reviewed through MRs. Additionally, the knowledge accumulation and recall mechanism provided by TeamAI CLI is a capability that cannot be achieved with local configuration methods.

Q: How does the knowledge accumulation trigger mechanism for friction signals work? How is accuracy ensured?
A: The tool detects friction signals during session interactions, such as interruptions, refusals, and retries. When the accumulated friction signals reach a predefined threshold, the tool prompts the developer to summarize the current pain points. After confirmation from the developer, the content is automatically anonymized and pushed to the team repository. Accuracy is primarily ensured through two mechanisms: one is the multidimensional comprehensive judgment of friction signals to reduce false triggers, and the other is the manual confirmation step before pushing, which provides the final quality check.

Q: Does using TeamAI CLI require the team to follow specific Git collaboration standards?
A: The tool relies on standard Git operations (branches, MRs, merges), but to use TeamAI CLI efficiently, it is recommended that the team follow basic Git collaboration standards, including: using branches for development rather than pushing directly to the main branch, configuring branch protection rules that require MR reviews before merging, and maintaining a clear repository directory structure with documentation. These standards ensure the quality and traceability of configuration changes and reduce the probability of conflicts during collaborative work.

Q: Will the knowledge recall feature leak sensitive team information?
A: TeamAI CLI includes a built-in privacy removal mechanism during the knowledge accumulation process. The experience content pushed to the team repository strips sensitive information from the session (such as keys or personal account details in code snippets). Additionally, the MCP and Hooks mechanisms allow for configurable security scan hooks to intercept sensitive data from being transmitted externally. The knowledge base itself is stored in the team's own Git repository, with access controlled by the team's Git repository permission system.

Q: Our team already has multiple tools each configured with their own set of rules. Can these be migrated to be managed uniformly by TeamAI CLI?
A: Yes, migration is possible. Existing rule files from each tool should be organized according to TeamAI CLI's directory specifications and then integrated into the team repository. TeamAI CLI's conversion layer supports converting the unified configuration format into the native format of each tool. It is recommended to first validate the conversion results in a test repository before officially switching. For personal configurations that were not previously managed by the team, you can selectively extract the parts that are common to the team and deposit them into the shared repository, while retaining individual-specific configurations in user-level settings.

9. Project Links

• GitHub Repository: https://github.com/Tencent/teamai-cli
• Tencent Open Source Organization Homepage: https://github.com/Tencent